Page 136 - Demo

Chapter 8134Modern learning techniques in tooth removalSummary of findingsIn Chapter 7 we present a classification model for tooth removal. Together with computer scientists, engineers and clinicians, brainstorm sessions were held in which we developed ‘features’ of tooth removal, based on the data presented in chapters 5 and 6. In total, we designed 75 features, such as ‘amount of rotation around the longitudinal axes or the ‘total amount of force delivered in the buccal direction . The importance of each feature to distinguish, for example, a molar from an incisor, was weighted in a feature selection step. It reduced the amount of features to 33. A Gaussian Naïve Bayes classification model was trained using the remaining features, with acceptable results. Especially so, given the expected variety in our dataset in combination with the relatively small data sample. On average and after 4-fold cross validation, the model was capable of determining the correct tooth classes in 86% (training set). In unseen data (test set), the average accuracy reached 54%. The model correctly classified the upper- or lower jaw, in 95% of the experiments. LimitationsAs stated previously, the sample size of our study is small and a high variety in our data is to be expected due to patient and anatomical factors. By including data from three different surgeons, the variety of our data is increased even further. This is beneficial for the ‘generalizability’ of our results, as a variety of techniques have been included. On the other hand, the heterogeneity of the data in combination with a small dataset makes it more difficult to train and interpret a classification model.Contributions to existing knowledgeThis study showed that the gathered data was of sufficient quality to use in a modern learning algorithm. It is a new way of looking at tooth removal procedures and opens doors to evidence based education in this field. For example, the degree of rotation around the longitudinal axis and the degree of rotation around the mesiodistal axis were selected for the model. From a clinical point of view, it is well known that these two rotations are essential in different tooth removal procedures. These metrics are also used in textbook instructions for tooth removal [9]. Tom van Riet.indd 134 26-10-2023 11:59